Digital multimeters (DMMS) and various other items of electronic test equipment typically have an internal Analog to Digital Converter (ADC) whose native voltage range is small compared to the overall range of use expected for the DMM. That is, the ADC may have a current driven virtual ground that produces a full scale output with a one hundred millivolt, one volt, or perhaps ten volt input. Modern DMMs generally use a resistive input network to produce from the applied voltage an input current that is within the range of the ADC.
This resistive input network resembles a classical voltage divider, in that it has two sections that are in series. In a classical voltage divider the input voltage is applied to one end while ground is at the other, with the divided output taken somewhere between, and is applied to some very high resistance measurement element, such as a FET or the grid of a tube. In a modern DMM the input is at one end of the resistive input network, a feedback voltage from an input amplifier that inverts its input is at the other, and a virtual ground is maintained at the division node (which is also the input to the input amplifier). The signal to be measured by the ADC is essentially the feedback voltage. It is fair to call this resistive network a voltage divider, since it is still a series combination of two resistances of a selected ratio, and the voltage drops across them will be in accordance therewith. It just so happens that it is part of a larger servo arrangement that keeps the voltage at the junction (the virtual ground) very near zero volts. We shall be content to refer to the resistive input network as simply a voltage divider.
The resulting ADC reading is then multiplied by a suitable scale factor (and perhaps otherwise processed) before being displayed as the DMM's result. The accuracy of the result then depends not only on the ADC, but also upon the voltage divider. This DC voltage divider often has a very high overall resistance, so as to maintain a suitably high input resistance for DC measurements.
Many DMMs also measure AC voltages in addition to DC, and do so using the same basic architecture, although the circuitry used might well be a different instance of that architecture. The need for a precision input voltage divider remains, as in the DC case, although the degree of precision is sometimes relaxed, owing to the difficulty in making precision AC measurements: the virtual ground now has to be an AC virtual ground, with everything that requires in terms of amplifier performance as a function of frequency. If one could be assured that the divider were entirely resistive, then things would be considerably easier. As it is, however, the designer has to contend with stray capacitances that variously shunt the resistive elements in the voltage divider. These strays exist across the resistive elements themselves, as well as exist from various nodes in the divider to other places, such as actual ground. At higher frequencies the decreasing capacitive reactances of the stray capacitances shunt the resistive elements and seriously disturb the division ratio, and thus destroy the accuracy of the measurement. The conventional method of coping with this situation is to add an additional capacitive voltage divider in parallel with the resistive one, with the corresponding nodes connected. That is, make a divider out of parallel RC sections that are connected in series. The RC time constants are all equal, and the added C's swamp out the strays. This type of structure is generally known as a compensated AC attenuator. The price paid is that the input impedance at high frequencies is now much less than before, although not so much as to become a serious issue.
In fact, while input resistances of 10 MΩ or (much!) higher are common for DC DMMs, a considerably lower value is often acceptable for AC measurements. A more important aspect of AC measurements is the flatness, or constancy, of the division ratio throughout a specified range of frequencies. As it turns out, capacitors are complex structures in their own right, and the effective ratio of a large capacitance to a smaller one may not stay the same over a wide range of frequencies. For example, they may have different dissipation factors. So, there is a limit, imposed by capacitor performance, to the extent that a resistive divider can be turned into a precision compensated AC attenuator. One way to reduce the need for additional capacitance in the attenuator is to use, for AC measurements, a different attenuator whose input resistance to the virtual ground is much lower, say, 1 MΩ. Such a lower value of resistance is less easily shunted by the stray capacitances, and a smaller added capacitance can be used to achieve compensation. This lesser need for compensatory capacitance eases the frequency response problem caused by capacitor performance (i.e., the capacitive divider's division ratio being a function of frequency as the different capacitors behave differently as frequency changes), but does not eliminate it entirely. The issue returns as the accuracy of the measurement increases. If the stray capacitances are swamped out, then performance issues in the additional compensatory capacitors become an issue, whereas if the strays are ignored, then even with reduced resistances those strays eventually become an issue, anyway, as frequency gets high enough. Thus, it seems that we are presented with a dilemma.
Now consider an input network for a precision AC voltage measurement of up to 1KV for frequencies of up to 1 MHZ. It should be not too big, not too expensive and of very flat frequency response, say to about 0.01%, while also being easy to trim (make minor adjustments for compensation) to maintain that flatness after installation in an actual circuit during manufacture. That is quite a wish list, and not easily achieved (the afore-mentioned dilemma) while using individual parts to form a conventional compensated attenuator. What to do?